Control of Inverse Response Process using Model Predictive ...utpedia.utp.edu.my/6148/1/FYP...

Control of Inverse Response Process using Model Predictive Controller

(Simulation)

by

Fawwaz bin Fuat

Dissertation submitted in partial fulfilment of

the requirements for the

Bachelor of Engineering (Hons)

(Chemical Engineering)

SEPTEMBER 2012

Universiti Teknologi PETRONAS

Bandar Seri Iskandar

31750 Tronoh

Perak Darul Ridzuan

I

CERTIFICATION OF ORIGINALITY

This is to certify that I am responsible for the work submitted in this project, that the

original work is my own except as specified in the references and acknowledgements,

and that the original work contained herein have not been undertaken or done by

unspecified sources or persons.

_____________________

FAWWAZ BIN FUAT

II

ABSTRACT

Model predictive control is an important model-based control strategy devised for large

multiple-input, multiple-output control problems with inequality constraints on the input

and outputs. Applications typically involve two types of calculations: (1) a steady-state

optimization to determine the optimum set points for the control calculations, and (2)

control calculations to determine the input changes that will drive the process to the set

points. The success of model-based control strategies such as MPC depends strongly on

the availability of a reasonably accurate process model. Consequently, model

development is the most critical step in applying MPC.

As Rawlings (2000) has noted, “feedback can overcome some effects of poor model, but

starting with a poor process model is a kind to driving a car at night without headlight.”

Finally the MPC design should be chosen carefully.

Model predictive control has had a major impact on industrial practice, with over 4500

applications worldwide. MPC has become the method of choice for difficult control

problems in the oil refining and petrochemical industries. However, it is not a panacea

for all difficult control problem(Shinkey, 1994; Hugo, 2000). Furthermore, MPC has had

much less impact in the order process industries. Performance monitoring of MPC

systems is an important topic of current research interest.

III

ACKNOWLEDGEMENT

Grateful……that’s the best word that I can say to appear my grateful after finished

completed final year project report at Universiti Teknologi PETRONAS. Firstly, I would

like to show my appreciation to my supervisor, Dr Lemma Dendena Tufa for supporting

me throughout this two semester and constantly provides me with assistance and help.

He is not only sacrificed their working time to supervise me to complete this project with

success but also taught me some moral virtues and how to behave a good employee in an

organization.

Secondly, I would like to thank the Chemical Engineering Department in

University of Technology PETRONAS (UTP) for giving me the opportunities to prepare

this technical report as my final year project. It’s has been help me a lot in refreshing

what I have learnt throughout five years I am here.

Furthermore, I would also like to express my thankfulness to my beloved parents

for giving me mentally supports to carry on this task.

Other than that, a special thanks to the reviewers of this report who contributed

significantly to project and to all of my friends, who offered valuable suggestion to

finished this report.

Finally, I would like to appear my hope that I can apply what I learnt when

finished my study soon.

IV

TABLE OF CONTENTS

CERTIFICATION . . . . . . . I

ABSTRACT . . . . . . . . II

ACKNOWLEDGEMENT . . . . . . III

CHAPTER 1: INTRODUCTION . . . . 1

1.1. Project Background . . . 1

1.2. Problem Statements . . . 1

1.3. Objectives and Scope of Study. . 2

CHAPTER 2: LITERATURE REVIEW . . . 3

CHAPTER 3: METHODOLOGY . . . . 16

3.1. Project Work . . 16

3.2 Research Methodology . . 18

3.3 Activities/Gantt Chart and Milestone . 19

CHAPTER 4: RESULT AND DISCUSSION . . 20

4.1 Data gathering and Analysis

4.2 Result and Discussion

CHAPTER 5: CONCLUSION AND RECOMMENDATION 45

5.1 Conclusion

5.2 Recommendation

REFERENCES . . . . . . . 46

APPENDICES

V

LIST OF FIGURES

Figure 1: Basic Structure of Model Predictive Controller

Figure 2: Model Predictive Controller block diagram

Figure 3: Project activities flow

Figure 4: MATLAB Simulation

Figure 5: Effect of Ƭi

Figure 6: Block diagram

Figure 7: Effect of Ƭi with Model Predictive Controller

Figure 8: Graph with Input rate weight Tuning

Figure 9: Graph with Output rate weight Tuning

Figure 10: Graph with Prediction Horizon Tuning

Figure 11: Graph with Control Horizon Tuning

Figure 12: Feedback Control Loop

Figure 13: Servo Control

Figure 14: Graphical Interpretation of IAE

Figure 15: Input (bottom) and output (top) data used for identification

Figure 16: The Bode plot of the actual model

Figure 17: Bode plot of the identified model and the actual model

VI

LIST OF TABLE

Table 1: Ghantt Chart

Table 2: Error with different Ƭ푖

Table 3: Error with default MPC setting with Ƭi Tuning

Table 4: Error with Input rate weight Tuning

Table 5: Error with Output rate weight Tuning

Table 6: Error with Prediction Horizon Tuning

Table 7: Error with Control Horizon Tuning

1

CHAPTER 1

INTRODUCTION

1.1 Project Background

Systems which have inverse responses are difficult to control. Many types of process

show inverse response over a part of the normal range operation among these is the

recycling system so common in engineering practice. Control can be greatly improved

by introducing a compensating feedback element into the control loop. In this project,

we will use Model Predictive Controller to control the inverse response processes.

1.2 Problem Statement

Some effect to the behavior of the process is affected by the tuning of the Model

Predictive Controller. The time taken to reach the set point and the error may be affected

by the parameter tuning for inverse response process. Based on the system, we have to

measure the error of the controller and make a comparison of efficiency between

response with and without using controller.

1.3 Objectives and Scope of Study

I. Analyze the effect of Ƭi response without using controller for a typical inverse

response process by MATLAB software.

II. Analyze the effect of Ƭi response by using Model Predictive Controller for a

typical inverse response process by MATLAB software.

III. Analyze the effect of input rate weight response by using Model Predictive

Controller for a typical inverse response process by MATLAB software.

2

IV. Analyze the effect of output rate weight response by using Model Predictive


V. Analyze the effect of prediction horizon response by using Model Predictive


VI. Analyze the effect of control horizon response by using Model Predictive


In order to study process control of inverse responses, the transfer functions should first

be developed. The transfer function which is the ratio of the output variable X to the

input signal Y in the Laplace Transformations is defines the performance of the process.

These equations represent a variety of typical combined processes or their

approximations. For this project, we will use second order equation of transfer function.

3

CHAPTER 2

LITERATURE REVIEW

The analysis, design, control, operation, and optimization of distillation columns

was studied and researched for long time ago [1,2]. As time goes by, engineers try to

come out with techniques and approaches to stabilize the control process due to the

present of nonlinearities and constraints. The most famous and popular method of model

predictive control is using the Dynamic Matrix Control (DMC) [3,4]. It is based on a

discrete time step response model that calculates a desired value of manipulated value

that remains unchanged during the next time step.

One of the applications of DMC is in multiple input and multiple output systems

(MIMO). In simulation experiment, MIMO is always compared in term of performance

with Single Input and Single Output (SISO) process system and the standard of model

predictive control. By this performance measurement, we can know which controller is

the most suitable for certain control process.

Besides, DMC introduced by Culter and Ramaker is available in most

commercial industrial distributed control systems and process simulation software

packages [5]. The three integral parts of any model predictive control algorithm which

are process model, the cost function and the optimization technique. There are various

forms of MPC such as Robust MPC that provides guaranteed feasibility and stability of

the process, feedback MPC that mitigates shrinkage of feasible region, pre-computed

MPC that using linear or quadratic programming, and decentralized MPC that can give

fast response.

4

Basic structure of MPC

The major contributions of MPC in control system industry are it handles

structural changes, easy to tune method, allow operation closer to constraint that can

increase the profit, can take account of actuator limitations, has plenty of time for on-line

computations, can handle non-minimal phase and unstable processes, and can handle

multivariable control problems naturally. The applications of MPC are always used in

distillation column, hydrocracker, pulp and paper plant, servo mechanism and robot arm.

The OBF models can be considered as a generalization of Finite Impulse

Response (FIR) models in which the filter are replaced with more complex orthonormal

basis filters [6,7]. There are many types of OBF such as Laguerre filter which has one

real pole for damped processes, Kautz filter which allows the incorporation of a pair

conjugate complex pole for modeling weakly damped processes, Generalized

orthonormal basis filter and Markov OBF which for a system involves time delay and

time delay estimation.

In many control application the desired performance cannot be expressed caused

by the process constraint [8,9]. Example for physical constraint is the actuator limits,

5

safety constraint such as temperature and pressure limits, performance constraint such as

overshoot and others constraints like manipulated variable constraint which hard limits

on input, manipulated variable rate constraint which are hard limit on the size of the

manipulated variable, and output variable constraint which hard or soft limits on the

output on the system are imposed to.

Based on my research, the advantages of using MPC are it is a straight forward

formulation based on well understood concepts, it can handle constraints, the

development time much shorter than for competing advanced control methods, easier to

maintain the changing model or the specs does not require complete redesign [10]. By

running different scenarios in linear and nonlinear simulations, you can evaluate

controller performance. You can adjust controller performance as it runs by tuning

weights and varying constraints [11].

In addition to developing more flexible control technology, new process

identification technology was developed to allow quick estimation of empirical dynamic

models from test data, substantially reducing the cost of model development [12]. More

importantly, MPC theory can lead to discoveries by which MPC can be improved

through new formulation [13]. This is important to make sure the efficiency of the

control process in minimizing the constraint and improve the control system in industry

or any manufacturing.

I. Model Predictive Control

Introduction to MPC

A general development is presented that gives great insight into the roles of both the

control algorithm and the process in the behavior of feedback systems. It also provides a

method for tailoring the feedback control algorithm to each specific application. Since

the model of the process is an integral part of the control algorithm, the controller

equation structure depends on the process model. Although the control algorithm is

different, the feedback concept is unchanged, and the selection criteria for manipulated

6

and controlled variables are the same. The algorithms could be used as replacements for

the PID controller in nearly all applications. PID controller algorithm is considered the

standard algorithm. Alternative algorithm is selected only when it provides better

performance.

MPC Structure

Consider the typical thought process used by a human operator implementing a feedback

control manually. The approach used by the operator has three important characteristics:

I. It uses a model of the process to determine the proper adjustment to the

manipulated variable , because the future behavior of the controlled variable can

be predicted from the values of the manipulated variable.

II. The important feedback information is the difference between the predicted

model response and the actual process response. If this difference is zero, the

control would be perfect, and no further correction would be needed

III. This feedback approach can result in the controlled variable approaching its set

point after several iterations, even with modest model errors

A continuous version of the approach can be automated with the general predictive

control structure.

7

The variable Emis equal to the effect of the disturbance, Gd(s)D(s), if the model is perfect

(Gm(s)=Gp(s)). The structure highlights the disturbance for feedback correction.

However, the model is essentially never exact. The feedback signal includes the effect of

the disturbance and the model error, or mismatch. The feedback signal can be considered

as a model correction. It is used to correct the set point so as to provide a better target

value, Tp(s), to the predictive control algorithm. The controller calculates the value of the

manipulated variable based on the corrected target.

The closed-loop transfer functions for the setpoint and disturbances are:

The controller algorithm, Gcp(s), for the predictive structure is to be determined to give

good dynamic performance. Let us determine a few properties of the predictive structure

that establish important general features of its performance and give guidance for

designing the controller. A very important control performance objective is to ensure

that the controlled variable returns to its set point in steady state. This objective can be

evaluated from the closed-loop transfer functions by applying the final value theorem

)()()(1)()(

)()()()()(1)()()(

)(SP)(CV

sGsGsGsGsG

sGsGsGsGsGsGsGsG

ss

mpcp

pcp

mspvcp

pvcp

)()()(1

)()()(1)()()()()(1

)()()(1)()(CV

sGsGsGsGsGsG

sGsGsGsGsGsGsGsG

sDs

mpcp

dmcp

mspvcp

dmcp

8

and determining whether the final value of the controlled variable, expressed as a

deviation variable from the initial set point, reaches the set point.

The next control performance objective is perfect control. Perfect control means the

controlled variable never deviates from the set point.

CV(s)/D(s) = 0 and CV(s)/SP(s) = 1 provide the basis for the following condition:

Perfect control can be achieved if the controller could be set equal to the inverse of the

process dynamic model. Block diagram algebra can be applied to derive the following

condition for the behavior of the manipulated variable.

The perfect control system must invert the true process in some manner.

)()( 1 sGsG mcp

)()(

1)()(1)()(

)()()(1)()(

)()(MV

sGsG

sGsGsGsG

sGsGsGsGsG

sDs

p

d

pcp

cpd

mpcp

cpd

9

II. Inverse Response

Inverse response behavior appears when the initial response of the output variable is in

the opposite direction to the steady state value. In chemical process industry this

phenomenon occurs in several systems such as distillation column and drum boilers [14].

The reason for the inverse response is that the process transfer function has an odd

number of zeros in the open right half plane. This non-minimum phase characteristics of

the process affects the achievable closed-loop performance because the controller

operates on wrong sign information in the initial time of the transient [15]. This fact

introduces essential limitations in terms of achievable output performance.

Let assume a process resulting from two parallel first-order stable processes having

opposite gain

Where K1, K2, τ1, and τ2 are positive constants. The overall transfer function of P(s) in

equation above can be posed as

Where

10

And

In this context, inverse response appears due to competing effects of slow and fast

dynamics. In concrete terms, it appears when the slower process has higher gain.

Therefore, the condition for inverse response reads as:

Study of Other Controller

Dynamic Matrix Control (DMC)

DMC is one of the most popular methods of model predictive control. A way to have

students explore the nature of DMC control is to use it on a simulated process [16].

DMC control is based on a discrete time step response model that calculates a desired

value of the manipulated value that remains unchanged during the next time step. The

new value of the manipulated variable the value that gives the smallest sum of squares

error between the set point and the predicted value predicted values of the controlled

variable. The dynamic model used to predict the future values of the controlled variable

is represented by a vector A, whose elements are defined as

11

Whrere

y(t) = the value of the controlled variable at time t

∆y(t0) = the change in the manipulated variable at t0

Thus, the response of a process to a step change ∆u, in the manipulated variable at t0

∆u(t0) is given by

Orthonormal Basis Control (OBF)

General orthonormal basis filter (OBF) models are generalization of FIR models,

presented a least square identification method to estimate a finite number of model

parameters based on Pulse, Laguerre and Kautz basis functions. They showed that the

form of the chosen basis functions and the accuracy of the poles used in them can

improve the identification results.

A test was simulated on the transfer function given and an output data was generated for

a given input data sequence with additive white noise of power 2.

12

An OBF model was developed using the input-output data with an initial time constant

τ=3, and time delay=0. It is observed that the resulting OBF model with only 5 OBF

terms has a very good accuracy.

Figure 15: Input (bottom) and output (top) data used for identification

The Bode plot of the actual model and that which is identified by the discussed

technique is presented below. As it can be seen, through there is small difference in the

time constants, and gain however, the Bode plots are nearly indistinguishable, hence

good enough for controller design purpose.

13

Figure 16: The Bode plot of the actual model

The response of the OBF model and the noisy output data plotted together for the same

input

Figure 17: The Bode plot of the identified model and the actual model

Internal Method Controller (IMC)

IMC controller is based on Brosilow (1979) and Garcia and Morari (1983). Since an

exact inverse is not possible, the IMC approach segregates and eliminates the aspects of

the model transfer function that make the calculation of realizable inverse impossible.

14

Thus, the first step to start the IMC approach is to factor the model into the product of

two facors.

The noninvertible part has an inverse that is not causal or is unstable

The steady state gain of this term must be 1.0

The invertible part has an inverse that is causal and stable, leading to

realizable, stable controller

The IMC Controller (idealized)

This design ensures the controller is realizable and the system is internally stable

Example of IMC Controller

To make the controller proper or semiproper, add a filter to make the controller

proper

For tracking setpoint changes,

Adjust the filter-tuning parameter to vary speed of the response of the closed-

loop system

)()()( sGsGsG mmm

)(sGm

)(sGm

1)()(~ sGsG mcp

3)15(039.0)(

s

sGm

3)15(039.0)(

ssGm 0.1)( sGm

039.0

)15()()(3

1

ssGsG mcp

)()()(~)(1 sGsGsGsG fmcpcp

nf ssG

11)(

15

: small --- response is fast

large --- the closed loop response is more robust (insensitive to model error)

Example:

The controller is proportional-derivative, which still might be too aggressive but can be

modified to give acceptable performance. Then, to make the controller semiproper,

115.10

039.01)()(

s

ssGsG cpf

)15.10(039.0)(

5.5

sesG

s

m )15.10(039.0)(

ssGm

sm esG 5.5)(

039.0

)15.10()()(~ 1

ssGsG mcp

1039.0)15.10()(

sssGcp

16

CHAPTER 3

METHODOLOGY/PROJECT WORK

3.1 Project work

Start

Research

Literature Review

Choosing Model Predictive Controller

Simulation

Result and Discussion

Final report

End

Figure 3: Project Activities Flow

Figure 4: MATLAB Simulation

17

The project is a simulation project. Specifically, it is a simulation of Model Predictive

Controller on how the performance to system. First and for most, the project will begin

with the research on several issues which had been mention in the research methodology

above.

With the collective information, the project will proceed with the literature review that

will mention about what people had done before for Model Predictive Controller

strategies and the performance based on certain process. Besides, the author will discuss

a basic knowledge of on how the Model Predictive Controller functioning and what are

the benefits when using Model Predictive Controller in a control system or process.

After completing the literature review, the further studies will move on to tune the

Model Predictive parameter of Controller approaches to make a performance

comparison between them due to a process stated in the objective of the proposal.

Besides, the author needs to identify the parameters that involved in measuring the

performance of each Model Predictive Controller due to system tested such as the value

Ƭi, input rate weight, output rate weight, prediction horizon and control horizon.

Then, the simulation of Model Predictive Controller will be done by using MATLAB.

After completing the simulation, the result and discussion will be done to know the

performance of each Model Predictive Controller based on the stated systems.

Lastly, all the studies and discussion will be compiled in the final report. Apart from

that, the justification of performance of the Model Predictive Controller also will be

mentioned in the discussion.

18

3.2 Research Methodology

Research is a method taken in order to gain information regarding the major scope of the

project. The sources of the research cover the handbook of condensate stabilization unit,

e-journal, e-thesis and several trusted link.

The steps of research:

1. Gain information of the Model Predictive Controller strategies and its

performance.

2. List down the parameter used to be tested for inverse response process.

3. Make a comparison between the parameter tuned based on a control process

decided.

Project Simulation

The project simulation will be done by using MATLAB. Then, we will make a

comparison between parameter tuned based on a system that we already choose early in

the topic.

19

3.3 Activities/Gantt Chart and Milestone

No Detail/Week 1 2 3 4 5 6 7

8 9 10 11 12 13 14 15

1 Project Work Continues

2 Submission of Progress Report

3 Project Work Continues

4 Pre- EDX

5 Submission of Draft Report

6 Submission of Dissertation (soft bound)

7 Submission of Technical Paper

8 Oral Presentation

9 Submission of Dissertation (hard bound)

20

CHAPTER 4

RESULT AND DISCUSSION

THE EFFECT OF Ƭi WITHOUT CONTROLLER

Ƭi = 2 Ƭi = 4

Ƭi = 6 Ƭi = 8

21

Comparison between Ƭi = 2, 4, 6, and 8

Figure 5: Effect of Ƭi

22

Overall process for Ƭi = 2, 4, 6, and 8

Figure 6: Block Diagram

Error for Ƭi = 2, 4, 6 and 8

CV Ƭi

Input variation

(Error)

Output Variation

(Error)

CV1 2 1.0672 15.982

CV2 4 1.1896 17.9701

CV3 6 1.374 19.9583

CV4 8 1.5585 21.9464

Table 2: Error of different Ƭi

23

DISCUSSION FOR THE EFFECT OF Ƭi WITHOUT CONTROLLER

For the inverse response of second order process with different value of Ƭi, we can see

the behavior of each graph changed with step value of 1. When we increase the value of

Ƭi, the inverse response become lower compared to smaller value of Ƭi. By using

MATLAB software, we calculate the error without using any controller in the process

and make comparison between them. Based on the graph and data calculated, we can say

that lower value of Ƭi is better and more efficiency to the process because it gives lower

value of error for the system although without using any model controller.

24

THE EFFECT OF Ƭi WITH DEFAULT MPC SETTING

Ƭi = 2

Ƭi = 4

25

Ƭi = 6

Ƭi = 8

Figure 7: Effect of Ƭi with MPC

26

DATA AND ERROR VALUE FOR THE EFFECT OF Ƭi WITH DEFAULT MPC

SETTING

Table 3: Error with default MPC setting with Ƭi tuning

ti=2 ti=4 ti=6 ti=8

5.6 0 0 0

5.52 0.08 5.6 5.6 1.4 1.4 1.4 1.4

3.69 1.83 6.34 0.74 0.998 0.402 1.05 0.35

1.95 1.74 5.47 0.87 0.646 0.352 0.722 0.328

0.875 1.075 4.14 1.33 0.558 0.088 0.628 0.094

0.422 0.453 2.72 1.42 0.579 0.021 0.634 0.006

0.371 0.051 1.43 1.29 0.612 0.033 0.657 0.023

0.505 0.134 0.371 1.059 0.634 0.022 0.676 0.019

0.679 0.174 -0.366 0.737 0.651 0.017 0.706 0.03

Total 5.537 -0.762 0.396 0.665 0.014 0.692 0.014

Total 13.442 Total 2.349 Total 2.264

IAE 6.5922 IAE 9.3456 IAE 10.4738 IAE 11.1559

ISE 6.125 ISE 11.4942 ISE 11.0621 ISE 12.5704

ITAE 21.5067 ITAE 35.3649 ITAE 50.6487 ITAE 54.489

27

DISCUSSION FOR THE EFFECT OF Ƭi WITH DEFAULT MPC SETTING

For this controller tuning, we set all parameter in default setting such as input rate

weight=0.1, output rate weight=1.0, prediction horizon=10 and control horizon=2. We

used Ƭi=2, 4, 6 and 8 for this tuning to see the behavior of the graph after it been

simulated and make comparison between the process without using any controller. After

the simulation was run, the error value of Ƭi=2 using MPC controller is lower that is

6.5922 compare to the process without controller which is 15.982. The different between

these two errors is 9.3898. it is same goes to Ƭi=4, 6, and 8. The error becomes lower

because the MPC can predict the past input or output to become a predicted output and

come out with future error. Then, the future error undergoes optimization process to

reduce the cost function and constraint and finally come out with future input which will

send back to the controller. Then, the MPC will send the corrective signal to the set point

to minimize the error. Thus, for second order process it is better to use small number of

Ƭi in the transfer function.

28

EFFECT OF MPC INPUT RATE WEIGHT= 0.1, 1, 10 TUNING FOR Ƭi = 6, OTHERS

ARE DEFAULT.

Rate Weight=0.1

Rate Weight=1.0

29

Rate Weight=10

Figure 8: Graph with Input rate Weight Tuning

30

Table 4: Error with Input Rate Weight Tuning

weight=0.1 weight=1 weight=10

0 0 0

1.4 1.4 -0.471 0.471 -0.00639 0.00639

0.998 0.402 -0.876 0.405 -0.0128 0.00641

0.646 0.352 -1.23 0.354 -0.0191 0.0063

0.558 0.088 -1.54 0.31 -0.0255 0.0064

0.579 0.021 -1.84 0.3 -0.0319 0.0064

0.612 0.033 -2.13 0.29 -0.0382 0.0063

0.634 0.022 -2.42 0.29 -0.0446 0.0064

0.651 0.017 -2.72 0.3 -0.0509 0.0063

0.665 0.014 -3.05 0.33 -0.0573 0.0064

Total 2.349 -3.41 0.36 -0.0636 0.0063

Total 3.41 Total 0.0636

IAE 10.4738 IAE 8.092 IAE 9.9656

ISE 11.0621 ISE 6.6703 ISE 9.9314

ITAE 50.6487 ITAE 37.6566 ITAE 49.7679

31

DISCUSSION FOR THE EFFECT OF MPC INPUT RATE WEIGHT= 0.1, 1, 10

TUNING FOR Ƭi = 6, OTHERS ARE DEFAULT.

For this tuning of input rate weight, we only use Ƭi=6 with different value of input rate

weight which are 0.1, 1.0 and 10. The default value of input rate weight is 0.1. Based on

the graph above, the increment of input rate weight value from 0.1 to 1.0 decreases the

error value for the controlled variable and decrease the move sizes significantly,

especially the initial move. The controller also minimizes the weighted sum of

manipulated variable deviations from their nominal values. But for manipulated variable,

the error is increase a bit only when the input rate weight is increase, then it decreases to

0.0636 when we increase the weight rate to 10. The value of input rate weight is affected

to both manipulated and controlled variable error. So it is important to tune the input rate

weight to ensure that the error is the least for manipulated and controlled variables

respectively.

32

EFFECT OF MPC OUTPUT RATE WEIGHT= 0.1, 1, 10 TUNING FOR Ƭi = 6,

OTHERS ARE DEFAULT.

Rate Weight=0.1

Rate Weight=1.0

33

Rate Weight=10

Figure 9: Graph with Output Rate Weight Tuning

output rate weight

weight=0.1 weight=1 weight=10

0 0 0

-0.471 0.471 1.4 1.4 3.47 3.47

-0.876 0.405 0.998 0.402 0.867 2.603

-1.23 0.354 0.646 0.352 0.852 0.015

-1.54 0.31 0.558 0.088 0.868 0.016

-1.84 0.3 0.579 0.021 0.878 0.01

-2.13 0.29 0.612 0.033 0.887 0.009

-2.42 0.29 0.634 0.022 0.895 0.008

-2.72 0.3 0.651 0.017 0.903 0.008

-3.05 0.33 0.665 0.014 0.91 0.007

-3.41 0.36 Total 2.349 Total 6.146

Total 3.41

34

Table 5: Error with Output Rate Tuning

DISCUSSION FOR THE EFFECT OF MPC OUTPUT RATE WEIGHT= 0.1, 1, 10


For this tuning of input rate weight, we only use Ƭi=6 with different value of output rate

weight which are 0.1, 1.0 and 10. The default value of output rate weight is 1. The

output weights let you dictate the accuracy with which each output must track its

setpoint. Specifically, the controller predicts deviations for each output over the

prediction horizon which we set it as default, 10. It multiplies each deviation by the

output's weight value. The weights must be zero or positive. A large weight on a

particular output causes the controller to minimize deviations in that output. Based on

the graph above, the increment of output rate weight value from 0.1 to 1.0 decreases the

error value for the controlled variable and decrease the move sizes significantly,

especially the initial move. The controller also minimizes the weighted sum of

manipulated variable deviations from their nominal values. But for manipulated variable,

the error is decrease a bit only when the input rate weight is increase, then it increases to

6.146 when we increase the weight rate to 10. The value of output rate weight is affected

to both manipulated and controlled variable error. So it is important to tune the input rate

weight to ensure that the error is the least for manipulated and controlled variables

respectively.

IAE 8.092 IAE 10.4738 IAE 10.5936

ISE 6.6703 ISE 11.0621 ISE 11.5651

ITAE 37.6566 ITAE 50.6487 ITAE 49.2717

35

EFFECT OF MPC PREDICTION HORIZON= 2, 10, 50 TUNING FOR Ƭi = 6,

OTHERS ARE DEFAULT.

Prediction Horizon= 2


36


PH=2 PH=10 PH=50

0 0 0

-5.25 5.25 1.4 1.4 5.95 5.95

-5.58 0.33 0.998 0.402 2.93 3.02

-5.48 0.1 0.646 0.352 1.25 1.68

-6.07 0.59 0.558 0.088 0.982 0.268

-7.25 1.18 0.579 0.021 1.05 0.068

-8.8 1.55 0.612 0.033 1.09 0.04

-10.6 1.8 0.634 0.022 1.08 0.01

-12.8 2.2 0.651 0.017 1.07 0.01

-15.3 2.5 0.665 0.014 1.06 0.01

-18.3 3 Total 2.349 Total 11.056

Total 18.5

Figure 10: Graph with Prediction Horizon Tuning

37

Table 6: Error with Prediction Horizon Tuning

DISCUSSION FOR THE EFFECT OF MPC PREDICTION HORIZON= 2, 10, 50


Prediction horizon is the number of control intervals over which the outputs are to be

optimized. Based on the data calculated, the error for the changes of prediction horizon

is not in sequence, means it is not increasing or decreasing based on the prediction

horizon number. When we change the prediction horizon from 2 to 10, the input error is

decreasing too much but the output error is increasing from 2.0298 to 10.4738. After

that when the prediction horizon is increase to 50, the input error is increasing and the

output error is increasing a bit only. So it is important to tune the prediction horizon

number to ensure that the error is the least for manipulated and controlled variables

respectively. Different tuning will give different behavior of input and output graph

which may affect the efficiency and time taken for the process to reach the set point.

IAE 2.0298 IAE 10.4738 IAE 10.6223

ISE 1.6624 ISE 11.0621 ISE 12.8529

ITAE 2.331 ITAE 50.6487 ITAE 44.7367

38

EFFECT OF MPC CONTROL HORIZON= 1, 5, 10 TUNING FOR Ƭi = 6, OTHERS

ARE DEFAULT.

Control Horizon= 1

Control Horizon= 2

39

Control Horizon= 10

Figure 11: Graph with Control Horizon Tuning

CH=1 CH=5 CH=10

0 0 0

-3.54 3.54 -3.29 3.29 -5 5

-3.65 0.11 -3.27 0.02 -5.27 0.27

-3.94 0.29 -2.97 0.3 -5.16 0.11

-4.48 0.54 -3.06 0.09 -5.68 0.52

-5.23 0.75 -3.45 0.39 -6.73 1.05

-6.15 0.92 -3.98 0.53 -8.12 1.39

-7.25 1.1 -4.58 0.6 -9.75 1.63

-8.53 1.28 -5.24 0.66 -11.7 1.95

-10 1.47 -5.96 0.72 -13.9 2.2

-11.7 1.7 -6.77 0.81 -16.5 2.6

Total 11.7 Total 7.41 Total 16.72

40

Table 7: Error with Control Horizon Tuning

DISCUSSION FOR THE EFFECT OF MPC CONTROL HORIZON= 1, 5, 10 TUNING

FOR Ƭi = 6, OTHERS ARE DEFAULT.

Control horizon sets the number of control intervals over which the manipulated

variables are to be optimized. Based on the data calculated, the error for the changes of

control horizon is not in sequence, means it is not increasing or decreasing based on the

control horizon number inserted. When we change the control horizon from 1 to 5, the

input error is decreasing too much but the output error is increasing from 4.3855 to

6.4548. After that when the control horizon is increase to 10, the input error is

increasing and the output error is decreasing to 2.5361. So it is important to tune the

control horizon number to ensure that the error is the least for manipulated and

controlled variables respectively. Different tuning will give different behavior of input

and output graph which may affect the efficiency and time taken for the process to reach

the set point.

IAE 4.3855 IAE 6.4548 IAE 2.5361

ISE 2.5723 ISE 4.5287 ISE 1.7408

ITAE 16.7406 ITAE 31.5615 ITAE 6.0361

41

Feedback Controller.

Feed forward controllers also exist but are more complicated to implement. Here we will

describe the use of feedback controllers.

The purpose of a servomechanism is to provide one or more of the following objectives:

I. accurate control of motion without the need for human attendants (automatic

control)

II. maintenance of accuracy with mechanical load variations, changes in the

environment, power supply fluctuations, and aging and deterioration of

components (regulation and self-calibration)

III. control of a high-power load from a low-power command signal (power

amplification)

IV. control of an output from a remotely located input, without the use of mechanical

linkages (remote control, shaft repeater).

In feedback control the variable required to be controlled is measured. This measurement

is compared with a given set-point. The controller takes this error and decides what

action should be taken by the manipulated variable to compensate for and hence remove

the error.

Figure 12: Feedback Control Loop

42

The advantage of this type of control is that it is simple to implement. Not only does the

feedback control system require no knowledge of the source or nature of the

disturbances, but it also requires minimal detailed information about how the process

itself works. Feedback control action is entirely empirical. So long as an adjustment is

being made in the correct sense then the control system should remove the effect of an

external disturbance.

A servo control loop is one which responds to a change in setpoint. The setpoint may be

changed as a function of time (typical of this are batch processes), and therefore the

controlled variable must follow the setpoint.

Figure 13: Servo Control

Response of second order process

Second order transfer function can arise physically whenever two first-order processes

are connected in series. For example, two stirred-tank blending processes, each with a

first-order transfer function relating inlet to outlet mass fraction, might be physically

connected so the outflow stream of the first tank is used as the inflow stream of the

second tank. Figure below illustrates the signal flow relation for such a process. Here

(ݏ)ܩ = (ݏ)(ݏ) =

2ܭ1ܭݏ) + ݏ2)(1 + 1) =

ܭݏ) + ݏ2)(1 + 1)

43

Calculating the Error

Controller tuning have been developed that optimize the response for a simple process

model and set point change. The optimum settings minimize an integral error criterion.

Three popular integral error criteria are:

1. Integral of the absolute value of the error (IAE)

ܧܣܫ = න ݐ|(ݐ)|∞

Where the error signal e(t) is the difference between the set point and the

measurement.

2. Integral of the squared error (ISE)

ܧܫ = න ݐଶ(ݐ)∞

3. Integral of the time-weighted absolute error (ITAE)

ܧܣܫ = න |(ݐ)|ݐ∞

ݐ

ܭ1 + 1

ܭ

2 + 1

X(s) U(s) Y(s)

Table 14: Graphical interpretation of IAE.

44

The ISE criterion penalizes large errors, while the ITAE criterion penalizes errors that

persist for long periods of time. In general, the ITAE is the preferred criterion because it

usually results in the most conservative controller settings. By contrast, the ISE criterion

provides the most aggressive settings, while the IAE criterion tends to produce controller

settings that are between those for the ITAE and ISE criterion.

45

CHAPTER 5

CONCLUSION

In conclusion, this project is to analyze the behavior of Model Predictive controller for

parameter tuning due to a system of inverse response control process. As the result of the

behavior measurement, we should have a basic knowledge on how to tuning the

controller for inverse response process. This is because inverse response is difficult to

control and the tuning may affect either the input or output error measurement. Besides,

by using Model Predictive controller is better than that not using any controller for any

process to minimize the error and increase the efficiency of the process.

46

REFERENCES

[1] HesamKomariAlaei, KarimSalahshoor, HamedKomariAlaei, 2010, Model Predictive Control of Distillation Column bases Recursive Parameter Estimation Method Using HYSYS Simulation. Page 308.

[2] Adrian Wills, March 2003, Doctor of Philosophy, Barrier Function Based Model Predictive Control.

[3] Charles R. Nippert, Simple Models That Illustrate Dynamic Matrix Control

[4] Sachin C. Patwardhan, Workshop on Model Predictive Control.

[5] RashmiRanjankar, AngshumanKalita, Dynamic Matrix Control

[6] D.T. Lemma, M. Ramasamy and M. Shuhaimi, 2010, System Identification using Orthonormal Basis Filters.Pages 2517.

[7] JieZeng, Raymond de Callafon, November 5-11, 2005, Filtersparametrized by Orthonormal Basis Functions for active noise control. Pages 1.

[8] Jan Maciejowski, Introduction to Predictive Control,

[9] Manfred Morari, Jay E. Lee, and Carlos E. Garcia, March 15, 2002, Model Predictive Control.pages 35.

[10] Glorinevsky, 2005, Lecture Note, The Concept of Model Predictive Control

[11] Model Predictive Control Toolbox, <http://www.mathworks.com/products/mpc/.

[12] Jan Maciejowski, Introduction to Predictive Control,

[13] Glorinevsky, 2005, Lecture Note, The Concept of Model Predictive Control

[14] F.G. Shinskey. Process control system.Mcgraw-Hill, 2nd. Edition,1979.

[15] S.Alcantara, C. Pedret, R. Vilanova, P. Balaguer and A. Ibeas, Systems Engineering and Automatic Control Group, 2008.

[16] Charles R. Nippert, Widener University, Simple Model that Illustrate DMC, 2002

APPENDICES

APPENDIX 1

MATLAB Coding

I. Input Variation Calculation

function [Iv]=inputvariation(MV) dif=filter([1 -1],[1],MV); absdif=abs(dif); Iv=sum(absdif);

II. Output Variation Calculation function IE=outputvariation(t,y,SP) %This function calculates the integral error IT1=trapz(t,y); IT2=trapz(t,SP); IE=IT2-IT1; III. Error Calculation

function IE=outputvariation(t,y,SP) %This function calculates the integral error IT1=trapz(t,y); IT2=trapz(t,SP); IE=IT2-IT1;

IV. Error Calculation for Model Predictive Controller

h = findobj(gcf, 'type', 'line'); xvalues = get(h, 'xdata'); yvalues = get(h, 'ydata'); p=cell2mat(yvalues(3));p=p'; q=cell2mat(xvalues(3));q=q'; ysp=1 e=abs(p-ysp); IAE=trapz(q,e) e2=e.^2; ISE=trapz(q,e2) e3=e.*q; ITAE=trapz(q,e3)

V. Transfer Function Ƭi = 2

>> Gp1=tf([-2 1],[4 1])

Gp2=tf([1],[10 1])

Transfer function:

-2 s + 1

--------

4 s + 1

Transfer function:

1

--------

10 s + 1

>> Gp=Gp1*Gp2

Transfer function:

-2 s + 1

-----------------

40 s^2 + 14 s + 1

>>

VI. Transfer Function Ƭi = 4

>> Gp1=tf([-4 1],[4 1])

Gp2=tf([1],[10 1])

Transfer function:

-4 s + 1

--------

4 s + 1

Transfer function:

1

--------

10 s + 1

>> Gp=Gp1*Gp2

Transfer function:

-4 s + 1

-----------------

40 s^2 + 14 s + 1

>>

VII. Transfer Function Ƭi = 6

>> Gp1=tf([-6 1],[4 1])

Gp2=tf([1],[10 1])

Transfer function:

-6 s + 1

--------

4 s + 1

Transfer function:

1

--------

10 s + 1

>> Gp=Gp1*Gp2

Transfer function:

-6 s + 1

-----------------

40 s^2 + 14 s + 1

>>

VIII. Transfer Function Ƭi = 8

>> Gp1=tf([-8 1],[4 1])

Gp2=tf([1],[10 1])

Transfer function:

-8 s + 1

--------

4 s + 1

Transfer function:

1

--------

10 s + 1

>> Gp=Gp1*Gp2

Transfer function:

-8 s + 1

-----------------

40 s^2 + 14 s + 1

>>

Control of Inverse Response Process using Model Predictive ...utpedia.utp.edu.my/6148/1/FYP...

Documents

Transcript of Control of Inverse Response Process using Model Predictive ...utpedia.utp.edu.my/6148/1/FYP...